cpu: add zero_denormals method

Add a method to enable/disable the denormals flush-to-zero and
denormals-as-zero CPU options.

Add a config option to make it possible to disable this again.

Fixes high CPU usage when dealing with denormals, which can happen
in many DSP functions.

Fixes #1681

spa/include/spa/support/cpu.h

@@ -113,7 +113,7 @@ struct spa_cpu { struct spa_interface iface; };
 struct spa_cpu_methods {
 	/** the version of the methods. This can be used to expand this
 	  structure in the future */
-#define SPA_VERSION_CPU_METHODS	1
+#define SPA_VERSION_CPU_METHODS	2
 	uint32_t version;
 
 	/** get CPU flags */
@@ -130,6 +130,10 @@ struct spa_cpu_methods {
 
 	/* check if running in a VM. Since:1 */
 	uint32_t (*get_vm_type) (void *object);
+
+	/* denormals will be handled as zero, either with FTZ or DAZ.
+	 * Since:2 */
+	int (*zero_denormals) (void *object, bool enable);
 };
 
 #define spa_cpu_method(o,method,version,...)				\
@@ -146,10 +150,12 @@ struct spa_cpu_methods {
 #define spa_cpu_get_count(c)		spa_cpu_method(c, get_count, 0)
 #define spa_cpu_get_max_align(c)	spa_cpu_method(c, get_max_align, 0)
 #define spa_cpu_get_vm_type(c)		spa_cpu_method(c, get_vm_type, 1)
+#define spa_cpu_zero_denormals(c,e)	spa_cpu_method(c, zero_denormals, 2, e)
 
 /** keys can be given when initializing the cpu handle */
 #define SPA_KEY_CPU_FORCE		"cpu.force"		/**< force cpu flags */
 #define SPA_KEY_CPU_VM_TYPE		"cpu.vm.type"		/**< force a VM type */
+#define SPA_KEY_CPU_ZERO_DENORMALS	"cpu.zero.denormals"	/**< zero denormals */
 
 /**
  * \}

spa/plugins/support/cpu-arm.c

@@ -123,3 +123,40 @@ arm_init(struct impl *impl)
 
 	return 0;
 }
+
+
+static int arm_zero_denormals(void *object, bool enable)
+{
+#if defined(__aarch64__)
+	uint64_t cw;
+	if (enable)
+		__asm__ __volatile__(
+			"mrs	%0, fpcr		\n"
+			"orr	%0, %0, #0x1000000	\n"
+			"msr	fpcr, %0		\n"
+			"isb				\n"
+			: "=r"(cw)::"memory");
+	else
+		__asm__ __volatile__(
+			"mrs	%0, fpcr		\n"
+			"and	%0, %0, #~0x1000000	\n"
+			"msr	fpcr, %0		\n"
+			"isb				\n"
+			: "=r"(cw)::"memory");
+#else
+	uint32_t cw;
+	if (enable)
+		__asm__ __volatile__(
+			"vmrs	%0, fpscr		\n"
+			"orr	%0, %0, #0x1000000	\n"
+			"vmsr	fpscr, %0		\n"
+			: "=r"(cw)::"memory");
+	else
+		__asm__ __volatile__(
+			"vmrs	%0, fpscr		\n"
+			"and	%0, %0, #~0x1000000	\n"
+			"vmsr	fpscr, %0		\n"
+			: "=r"(cw)::"memory");
+#endif
+	return 0;
+}

spa/plugins/support/cpu-x86.c

@@ -176,3 +176,15 @@ x86_init(struct impl *impl)
 
 	return 0;
 }
+
+static int x86_zero_denormals(void *object, bool enable)
+{
+	unsigned int mxcsr;
+	mxcsr = __builtin_ia32_stmxcsr();
+	if (enable)
+		mxcsr |= 0x8040;
+	else
+		mxcsr &= ~0x8040;
+	__builtin_ia32_ldmxcsr(mxcsr);
+	return 0;
+}

spa/plugins/support/cpu.c

@@ -78,11 +78,14 @@ static char *read_file(const char *name, char *buffer, size_t len)
 # if defined (__i386__) || defined (__x86_64__)
 #include "cpu-x86.c"
 #define init(t)	x86_init(t)
+#define impl_cpu_zero_denormals x86_zero_denormals
 # elif defined (__arm__) || defined (__aarch64__)
 #include "cpu-arm.c"
 #define init(t)	arm_init(t)
+#define impl_cpu_zero_denormals arm_zero_denormals
 # else
 #define init(t)
+#define impl_cpu_zero_denormals NULL
 #endif
 
 static uint32_t
@@ -197,6 +200,7 @@ static const struct spa_cpu_methods impl_cpu = {
 	.get_count = impl_cpu_get_count,
 	.get_max_align = impl_cpu_get_max_align,
 	.get_vm_type = impl_cpu_get_vm_type,
+	.zero_denormals = impl_cpu_zero_denormals,
 };
 
 static int impl_get_interface(struct spa_handle *handle, const char *type, void **interface)
@@ -265,6 +269,8 @@ impl_init(const struct spa_handle_factory *factory,
 			this->flags = atoi(str);
 		if ((str = spa_dict_lookup(info, SPA_KEY_CPU_VM_TYPE)) != NULL)
 			this->vm_type = atoi(str);
+		if ((str = spa_dict_lookup(info, SPA_KEY_CPU_ZERO_DENORMALS)) != NULL)
+			impl_cpu_zero_denormals(this, spa_atob(str));
 	}
 
 	spa_log_debug(this->log, "%p: count:%d align:%d flags:%08x",

src/daemon/pipewire.conf.in

@@ -15,6 +15,7 @@ context.properties = {
     #mem.mlock-all                         = false
     #clock.power-of-two-quantum            = true
     #log.level                             = 2
+    #cpu.zero.denormals                    = true
 
     core.daemon = true              # listening for socket connections
     core.name   = pipewire-0        # core name and socket name

src/pipewire/context.c

@@ -395,9 +395,14 @@ struct pw_context *pw_context_new(struct pw_loop *main_loop,
 			pw_properties_update_string(properties, str, strlen(str));
 		pw_properties_set(properties, "vm.overrides", NULL);
 	}
-	if (pw_properties_get(properties, PW_KEY_CPU_MAX_ALIGN) == NULL && cpu != NULL)
+	if (cpu != NULL) {
+		if (pw_properties_get(properties, PW_KEY_CPU_MAX_ALIGN) == NULL)
 			pw_properties_setf(properties, PW_KEY_CPU_MAX_ALIGN,
 				"%u", spa_cpu_get_max_align(cpu));
+		if ((str = pw_properties_get(properties, SPA_KEY_CPU_ZERO_DENORMALS)) == NULL)
+			str = "true";
+		spa_cpu_zero_denormals(cpu, spa_atob(str));
+	}
 
 	if (getenv("PIPEWIRE_DEBUG") == NULL &&
 	    (str = pw_properties_get(properties, "log.level")) != NULL)